PrismML, an AI Lab established by researchers at Caltech and guided by Ion Stoica from UC Berkeley, has developed a compact version of its language models designed for smart glasses utilizing Qualcomm’s Snapdragon chips.
During Qualcomm’s Snapdragon Summit on Wednesday, the company highlighted PrismML’s innovative 1-bit Bonsai language model, which can operate directly on AI smart glasses powered by the Snapdragon AR1 Gen 1 Platform.
As previously highlighted, PrismML is notable for its ability to significantly reduce the size of larger AI models (in this instance, by 4 times) while maintaining nearly all of their performance in standard evaluations. This version for smart glasses consists of a model with 2 billion parameters, specifically optimized for understanding both visual and linguistic input, enabling wearers to inquire about their surroundings in real time.
PrismML’s broader mission revolves around developing open-weight AI that operates on various devices, optimizing the existing computational capabilities. The startup promotes this approach as a viable alternative to relying on the privacy assurances provided by proprietary AI companies, which are constantly in need of increasing computational resources.
The introduction of a model specifically designed for Qualcomm’s chip marks progress towards this ambition. However, no specific smart glasses integrating PrismML technology have been announced as of yet.
